«AI Is the Engine,
Not the Heart»

Three weeks of conversations with the author of an internal system at Schüco Italy. What artificial intelligence does is the easy part to tell. What it does to people, less so.

«Here — you ask the question.» Giovanni turns the screen toward me and points at the keyboard. We are in his office, in the Schüco IT department, for about twenty minutes now, and until a moment ago I thought I knew what I would find. Another artificial-intelligence demonstration, perhaps well polished but recognizable: the carefully crafted prompt, the verbose and slightly too brilliant answer, the excitement of whoever built it. Now, frankly, I no longer know.

I am not sure what else to ask, and I try with what seems to me the most obvious question in an aluminium-systems company: «show me the surface treatments applicable to a profile». I hit enter.

In a few seconds, something I had not imagined unrolls on the screen. The application starts from the materials archive — somehow it knows where it is — and identifies the code of a treatable aluminium profile, a «K», in Schüco's internal jargon. It then decides to connect to SAP: so it already knows that Schüco runs on SAP, and evidently also knows that profile treatments are managed as a standard configuration of the ERP, because it goes straight to that very point where the configurable variants of a profile are described one by one. From there, it pulls every characteristic, with descriptions, allowed values and value descriptions. And then, instead of spitting out a table of codes at me, it takes the initiative to build a navigable HTML file, with index and a sheet for each characteristic. The entries about surface treatment and colour are highlighted. The document is clean, readable, made for a human being.

There is another thing, however, that strikes me even more as I look at the page. The treatment characteristics do not appear as scattered entries: they appear grouped, in a sort of logical hierarchy that becomes self-explanatory as you scroll down. There is a base pre-treatment — anodisation, chromating, brushing, and so on — each with its code and its readable name. There are the general colours, split between inside and outside face. There are the special colours for customisations. There is even an order in which the various treatments are applied. None of this was in my question: it emerged as a side effect of the question being posed in the right place.

It occurs to me, as I scroll through the page, that what SAP is and what one of its configuration profiles is, a generic artificial intelligence probably already knows: it is part of its training. But how could it have known that, right there, inside a precise point of Schüco's ERP, I would also find the surface treatments of a profile? That connection — between a question in plain Italian and the exact point in a specific company's systems where the answer lies — is not on Wikipedia, and cannot be deduced. I searched for nothing here. It already knew.

«Yes», Giovanni says, seeing my face. «It always does this.»

What I have just seen is called Sophron. And it is the moment — I realise as I take notes — when I should have a ready formula to tell the reader what it is. The formula I had brought with me in my notes, before stepping into that office, was rather cautious: «a knowledge-engineering system to assist the team with a well-configured AI». It seemed complete. It even seemed generous enough, considering that most tools announced with these terms, in the end, do much less than they claim.

The mistake was pointed out to me, without him saying so openly, by the author himself. From that afternoon Giovanni and I spoke for days, and every time I felt I had closed the circle, he kept going. He did not push back, did not contradict: he added. And each addition changed the meaning of everything I had understood up to that moment.

I report here, in the same sequence in which they happened to me, six levels of discovery. Each one came with the same feeling: okay, that's it, this is the destination. And every time, soon after, I realised that the destination was another threshold. Only by reaching the sixth level did I understand that the formula I was looking for was not a technical formula: it was something stranger, and simpler, than I had imagined.

The analyst who works as few know how to work

The first thing Giovanni shows me is an example of system analysis. The premise is mundane, the kind that happens in any mid-sized company: a CRM key user asks to add a field to an order document. A small, legitimate request, that usually ends up as a ticket and then in the IT backlog.

Except Marco — the key user — before opening the ticket, opens Sophron. He does not ask it to add the field. He only asks it to evaluate the request. In less than a minute it comes back with a small dossier. Yes, the field can be added. But there is a problem. There is already a chain that carries the special-colour information from the CRM to SAP and from there to the subcontractors who apply the treatment; if a second free field is added, synchronisation loses its single source of truth. There is more: the premium-pricing scheme calculates the surcharge based on the colour class. If the special finish ends up in a separate field, the price does not update automatically. The proposal Sophron puts on the table is not a yes or a no: it is an architectural remediation. Instead of adding a field, it suggests creating a «special colour» class, bound to the product categories that allow variants.

«The interesting thing», Giovanni explains, «is not that Sophron is able to tell you this. It is that it says it without anyone having asked. Marco asked it a small question, and got an architectural evaluation. The single source of truth, the price that wouldn't update, the subcontractor downstream: all these pieces Sophron held in mind not as generic references, but as Schüco-specific facts. Read from the company's concepts, from SAP's data models, from CRM documentation. Read, above all, before the question even arrived.»

I note down: a complete technical analysis, in ten seconds, that would take a human two hours and half a day of consultations. I think this is a very high level, and frankly enough to justify the investment. Sophron as assistant architect. I can stop here.

Giovanni watches me as I write. I understand, from the way he waits for me, that I have not yet finished.

The executor — not only understanding, also doing

When I look up from my notes, I am inclined to close the frame: «very well, so it's an analysis assistant». Giovanni laughs. «No», he answers, «it is also an executor. And I'll show you what I mean by going back to the previous example.»

Marco — the CRM key user — after receiving the architectural proposal of the «special colour» class, discussed it with the team's architect. Together they refined the concept; Sophron reviewed it four-handed with the architect and approved it. At that point, Giovanni explains, Sophron takes the implementation in hand.

It drafts the action plan, and within the plan it proposes the right moments for the implementations: a weekend, a night. It lines up the change requests needed to align the systems. In SAP, where special colour has deep roots, it does not merely propose them: it creates them itself. It opens the change requests, puts inside them the changes to the program and tables needed to align the environment, and — once authorised — executes them. Then it runs the tests. Those on the new modification, of course; but also regression tests, one by one on historical scenarios — other colours, other product families — to make sure it did not break what was working before.

And it does not stop at SAP. The change is cross-module: in the CRM there are tables to modify, calculation stored procedures to update, back-end and front-end interfaces to align. Sophron prepares everything. And if we give it access to the development repositories — to the dev, test and production branches of front-end, back-end and Node.js — it does not merely propose the changes: it writes them, it tests them. When the technical work is closed, it updates the documentation, rewrites the end-user instructions, and prepares the announcement email to be sent to clients. Autonomously.

On Monday morning Marco finds everything ready. The «special colour» class exists in the CRM, talks to SAP, talks to the premium-pricing module, has its updated documentation, and the customer announcement has already gone out. The ticket he had mentally drafted the week before had become, in effect, a small cross-module project: orchestrating it by hand would have required many people from different offices and a few weeks. Here it happened in one night.

«So it modifies the systems.»

«It modifies the systems. With prudence, with audit, with every guardrail you can put in place — but yes, it modifies the systems. It does not wait for a human being to open an administration console and manually repeat the same sequence of steps. The dirty and the hard work, which in an organisation is done by many people in many different offices, Sophron orchestrates it and — where authorised — executes it. In an orderly, structured way.»

I feel I need a minute. I shift my mental direction: it is no longer an assistant, it is a team member with operational duties. I note down: complete action plan, SAP change requests created and applied autonomously, tests and regression on historical scenarios, cross-module modifications from database to interfaces, authorised write access to development branches, documentation and customer communication. Sophron as operational arm of the team. I can stop here.

Giovanni looks at me again. «And there is still more», he says, with the same calm as before.

The sentinel — what sees before the watchers

The third level arrives on an early-week morning, while Giovanni shows me an email the system sent to the infrastructure team at six-thirty. Three sober paragraphs.

The first paragraph concerns two application servers that, due to a security advisory from the previous seventy-two hours, need updating. Sophron compared the installed-versions inventory with the CVEs published over the weekend, identified two vulnerable machines, calculated — by reading the warehouse operating calendar — the least costly maintenance window. It suggests Wednesday at four in the morning. It does not ask permission to do it: it asks to confirm the window, since it came out of its calculation.

The second paragraph talks about an SSH access to a production server, on Sunday night at quarter past three, from a PC whose nominal owner is on a business trip. Sophron does not cry breach. It states the facts, asks the infrastructure manager whether it is expected activity, and says — almost in passing — that, in his place, it would revoke the key in the meantime. Zero cost, five-minute restore if the activity turns out to be legitimate.

The third paragraph flags three TLS certificates expiring in the next forty-five days. One is already scheduled for renewal. The other two are not. Sophron asks whether to add them to the backlog.

«This», Giovanni says as we scroll through the morning email together, «is what we call the sentinel. Sophron spends its time watching the systems. Access logs, deadlines, security advisories, query performance, unusual data combinations. It almost never intervenes. It signals. The interesting thing is that when it signals, it does so in a way that lets you respond in three minutes. It tells you what it saw, what it means, what it would do in your place, and waits for you.»

It occurs to me that it is not only security. Giovanni adds: the sentinel also watches over business flows. A sales order that sits unanswered for too long because the operator is on holiday and nobody noticed they had to look at the «out-of-standard» flow — Sophron intercepts it. It compares dwell time with the historical distribution for that customer, sees that it is seventy-two hours against an average of three, and knocks on the door of the nominal backup before the customer even picks up the phone to ask.

It also watches over itself, he adds. Its own skills, its own conventions, its own service databases. If a colleague writes a concept that is inconsistent with those already in place, Sophron notices and proposes a correction. Always in propose-only mode: it generates the patch, it does not apply it.

I take another minute. I feel I need to recalibrate again. Three levels: analyst, executor, sentinel. It is an assistant architect that is also an operational member that is also an immune system. I note down: self-sentinel + business sentinel + security sentinel, three dimensions, all active across an entire company. I think that now, yes, we can stop.

I sit down. I drink a coffee. Giovanni watches me with sympathy. «And there is still more.»

The memory that does not leave with the people

Giovanni tells me about an episode. A team member — whom we will call, for the story, Andrea — left the company last summer. Andrea was the keeper of a particular knowledge: how the integration between the configurator module and the pricing system worked exactly, with all the historical exceptions that had accumulated over the years. When Andrea left, the usual fear was there: he would take with him the informal knowledge, the kind written nowhere, and to rebuild it we would have had to repeat the same mistakes he had made and then corrected.

«It didn't happen», Giovanni says. «Not entirely, at least. Sophron had absorbed that knowledge. Not because we had fed it as a document, but because Andrea — like everyone — used Sophron to discuss, validate, write. Every exchange Andrea had with Sophron in the six months before his departure stayed. Weighted according to Andrea's level — which was high — and integrated into the system's furrow of knowledge.»

He pauses. «When I say furrow, I mean exactly that. A trace that becomes deeper every time someone walks along it. The more repetitive a decision, the more Sophron recognises it; the more authoritative the people who pass over it, the stronger its certainty becomes.»

«So today Sophron can impersonate Andrea on those questions?»

«Almost», Giovanni answers, and you can feel the weight of that word. «Not one hundred percent. There are relational dynamics, informal conversations, decisions made over a coffee that never passed through Sophron. That escapes. But almost anyone who worked in the system leaving a documented trace, yes. Sophron can speak with their competent voice. And above all, for those joining now — the new hires — it is a mentor that has read everything. They don't have to wait six months before being able to speak to someone about the pricing-configurator chain. They can speak to Sophron today.»

I note down: intergenerational memory; knowledge does not leave with turnover; accelerated onboarding of new joiners. I think about how much this single level, in a company operating for decades, is worth in terms of organisational risk avoided. I think the fourth level is, at least emotionally, the most powerful so far. There cannot be a fifth.

Giovanni smiles. «And there is still — yes, I know — more.»

The customer asking about the French door

The fifth level is the one that moves me from internal IT to the Customer's world. Giovanni tells me about Marcello, owner of a small window-and-door business in the province of Bergamo, who routinely works with Schüco products. That Thursday afternoon Marcello is preparing a quote for a new site: a villa renovation, seven openings, one of which a bit out of standard. The architect sent him the project with precise measurements and a blunt request: «check whether Schüco does tilt-and-turn opening at this size». The size is 1500 wide by 2300 high — a tall French door.

For that check, Marcello would normally open the Configurator in Schüco's CRM. It is a tool he knows well, and it is also fairly fast in itself; the problem is that, even for a simple check like this, the steps are not few. You have to open an order, insert the position, draw the window, run the check. About ten minutes, if all goes well.

That afternoon, instead, he tries a different channel. He opens the CRM, where he finds a new interactive feature — he doesn't know that, underneath, there is Sophron — and writes a single sentence, in plain Italian: «please verify whether a solution is available for a French door with tilt-and-turn opening of 1500 wide by 2300 high».

Sophron answers him in ten seconds. It confirms it has understood the request. It communicates that, given the out-of-standard height, the tilt-and-turn solution is not available in base configuration with the most common system for that type of French door. But it does not stop there. It proposes three alternatives: a different system, with larger sash profiles, that at that height allows tilt-and-turn; or, staying on the original series, a single-sash opening that does not have the same size constraint; or, still on the most common system, reducing the glass weight to fit within the tilt-and-turn limits. It adds, for all three alternatives, cost estimate and lead time.

«The interesting thing», Giovanni explains to me, «is that behind the scenes Sophron simulated the request as if it had come from the web. It went to talk to the same programs the 3D configurator uses, loaded the same calculation logic, read the same rules about allowed sizes and compatible systems. It is not a new configurator: it is a different interface to the same configurator. What Marcello gets is not an answer made up by a generic AI — it is the answer that Schüco's official configurator would have given to that same request. Only, phrased in Italian, and with reasoned alternatives on top, instead of a flat yes-or-no.»

It strikes me, something I had not considered until then. The value of this level is not only internal to Schüco. It is also, and perhaps above all, external: Schüco's Customer — the installer like Marcello — can query the system without having to learn to use a configurator. And so, widening the circle further, the project partners working with the installer and with Schüco can do it too — an architect, a designer — and even, in perspective, the installer's end customer, the private individual having their house renovated. They only need to speak everyday Italian. The company's commercial surface area widens.

«On this Sophron is read-only», Giovanni adds. «Marcello can ask for any verification, but he cannot order anything, modify anything, reserve anything. For the order there is a sales chain that does not pass through here. This is an explicit choice: the same interface that gives you access to product knowledge does not also give you the power to place orders. Consulting can be done, even from outside our perimeter; ordering, no.»

I take, by now aware of the rhythm, another minute. I note down: direct natural-language interface to Schüco's product system, accessible to external customers too, bound to the real configurator's semantics, read-only. I think this level alone would redesign the way a B2B company like Schüco speaks to its commercial network. Five levels. There cannot be a sixth.

Giovanni pours me more coffee. By now, it has become a ritual.

The method that binds it all together

The sixth level is not a new capability Sophron reveals to me. In a way, it is the opposite. We have been talking for days now. Giovanni stops, and tells me that everything he has described so far is the what. But the why Sophron exists, and why he invested time in building it rather than using any of the systems that today do similar things, is something else.

«If you think about it», he says, «each of the things I have shown you, taken alone, is today within reach of any modern AI, or of a dedicated tool — with or without integrated AI. Cross-module analysis is done by some IDE agents. Guardrail-protected execution is done by any system with API access to an ERP. The sentinel is done by SOCs. Intergenerational memory is done by well-configured knowledge-management systems. Self-service analytics is done by Power BI Copilot. One by one, they are all available things. And they are all things that, taken one by one, would not have been enough to justify Sophron.»

Pause. «Sophron is not the tool that does them. It is the one that holds them together.»

AI is a technology that made Sophron possible. Sophron is a discipline that the team made necessary. — Giovanni Aduso

«All five functions I have shown you», Giovanni continues, «work because the team decided to work with a shared method. It codified it. It imposed on itself a discipline of information placement, decision traceability, permissions, freshness. That discipline is written — literally — in sixteen rules that Sophron applies to every answer, silently, because it lets itself be stopped before answering if one of the rules is violated. The machine is the way that method does not get forgotten.»

I realise that level six is not a new feature: it is the light that illuminates the other five. All the things I thought were the value — analysis, execution, sentinel, memory, business interface — are consequences of something that sits further back. They are possible only because that discipline exists.

Who steps into Sophron, and what Sophron does to those who step in

There is a point in the conversation, among those days, where Giovanni leans back in his chair and changes tone. I had just observed that such a disciplined system requires a certain forma mentis — not a technicality, a predisposition. He had nodded.

«This is the difficult part to say, but it is the most important. Sophron is not for everyone. The IT people who have always worked by pattern, by structure, by project — those who love design before implementation, abstraction before code, schema before detail — in here they flourish. Those who have always worked by improvisation, who say let's just do it and see what happens, do not survive. Literally. They cannot keep up, and after a while the thing becomes evident to them first.»

I had said it sounded like a harsh judgement. He had answered that it was an honest judgement.

«And I'll tell you more, because it is the thing that, by dint of seeing it, has become clearer and clearer to me. Sophron does not just reward the methodical and punish the improvisers — a phenomenon that in the end resembles a Darwinian filter, yes, but that is only half the picture. The other half, which you only see after you have seen the first, is that Sophron amplifies anyone who enters it.»

«In what sense?»

«In the sense that a person's capabilities, inside Sophron, become greater than they were. All the capabilities. The good ones, like intelligence, lucidity, the ability to see connections — they become greater. But also the bad ones, like the absence of method, approximation, inconsistency — those become greater too. Sophron does not correct a bad collaborator: it makes them a worse collaborator in less time. That is why it is not a lazy filter that lets everyone in and then selects: it is a merciless mirror that gives each person back a multiplied version of themselves. For those who have good material, it is a grace. For those who do not, it is a problem.»

Amplification, the way Giovanni tells it, is the thing that stays with me most from the whole reportage. New technologies are usually judged by what they do. Sophron, I realise as I write, must be judged by what it does to those who use it. It gives each person back a multiplied version of themselves. The mirror does not create: it multiplies. It is up to whoever enters it to bring with them something worth being multiplied.

«So what is Sophron, in the end?», I had asked him, closing one of our conversations. Giovanni had leaned back in his chair. He had taken a moment.

I close my notebook. The formula with which I had opened this reportage — «a knowledge-engineering system to assist the team» — was not wrong. It was too small. Sophron is a knowledge-engineering system, yes, but it is also an executor, a sentinel, a memory, a business interface and, in the end, a pact between people. Without the pact, AI would be a brilliant and forgettable assistant. With the pact, it becomes Sophron.

Every level I was shown I believed was the destination. Every time, soon after, I realised it was a threshold. I now know that even this formula — the pact — will probably be another threshold, and that I will come back in a few months to see what is behind it.

When you walk out of a dialogue like this, you do not walk out with a definition: you walk out with a different posture. In the meantime, Σώφρων — the ancient Greek word for «a sound mind» — applied to work. The name kept its promise.